Zhuang Ziyi

School of Artificial Intelligence, Lishui University

Abstract：

This study focuses on the evolution and prospect of flood disaster prediction methods from traditional models to intelligent algorithms, addressing disaster prevention needs amid climate change and urbanization. Global floods are frequent, causing substantial economic and social losses in China, with existing technologies facing prominent bottlenecks: traditional hydrological-physical models adapt poorly to urbanization-driven underlying surface changes, while intelligent algorithms lack interpretability. Using 1990-2024 Chinese and English literatures, this study employs literature review and comparative analysis to systematically explore three method categories: traditional models (e.g., SWAT, HBV, SWMM), intelligent algorithms (e.g., LSTM, CNN), and hybrid models (physical mechanism-data-driven integration). It quantifies key indicators (accuracy: NSE/RMSE; efficiency: simulation runtime; robustness: data scarcity adaptability) and extracts the "mechanism-driven → data-driven → hybrid-driven" evolutionary law. Innovatively, it analyzes methods from the dual perspective of "technical core (physical mechanism/data fusion) - applicable scenario (urban short-term waterlogging/rural long-term flood, gauged/ungauged basins)" and constructs a preliminary "scenario-method matching framework", addressing the lack of scenario-specific analysis in existing reviews. This review provides theoretical references for flood prediction model optimization and smart water conservancy practice, supporting regional flood disaster prevention and mitigation decision-making.

Key Words：

flood disaster; prediction methods; physical models; machine learning; deep learning; hybrid models